Qq Infoviz

Part of Claude Code operator skills: a collection of skills for running a real Claude Code setup.

Infoviz: branded data visualization skill

Generate editorial-quality LinkedIn post images (1200x627) using HTML/CSS rendered via Playwright.

Design philosophy: data IS the point. Every pixel communicates information. Exaggerated sizes, big beautiful serif numbers, tinted labels. Think editorial magazine, not data science notebook.

Pipeline

1. Get the data: pull from your own data source (see below), use data provided in conversation, or derive it from context
2. Pick a chart type: bars, grid, columns, newspaper, or quote+stat
3. Generate HTML: write a self-contained HTML file to /tmp/infoviz-chart.html
4. Screenshot: python3 ~/.claude/skills/qq-infoviz/screenshot.py /tmp/infoviz-chart.html /tmp/chart-linkedin.png 1200 627 (pass width/height explicitly: the script defaults to a 1800x1800 square, which doesn’t match the 1200x627 canvas this skill designs for)
5. Show the result: open /tmp/chart-linkedin.png in your OS’s image viewer (open on macOS, xdg-open on Linux) so a human actually looks at it. Don’t just read the file as terminal text.
6. Iterate: adjust and re-screenshot

Brand system

Canvas

• Dimensions: 1200x627px (LinkedIn optimal 1.91:1)
• Background: #0f172a (dark navy / slate-900)
• Padding: 48px left/right (symmetric, both sides must match)

Fonts

<link href="https://fonts.googleapis.com/css2?family=Source+Serif+4:opsz,wght@8..60,300;8..60,400;8..60,600;8..60,700&family=Inter:wght@400;500;600;700&display=swap" rel="stylesheet">

Typography scale (converged through 5 rounds)

Color palette

Stat     Number    Accent    Label (tinted)
Red      #f87171   #ef4444   #fecaca
Orange   #fb923c   #f97316   #fed7aa
Amber    #fbbf24   #eab308   #fef08a
Blue     #60a5fa   #3b82f6   #bfdbfe
Purple   #a78bfa   #8b5cf6   #ddd6fe
Indigo   #818cf8   #6366f1   #c7d2fe
Teal     #2dd4bf   #14b8a6   (derive)
Green    #4ade80   #22c55e   (derive)

Fill patterns (4 winners from pattern battle)

All fills go bottom-up. Pattern applies only to the filled area, not the entire cell.

/* 1. diag, standard diagonal stripes */
background: repeating-linear-gradient(-45deg,
  rgba(R,G,B,0.12) 0px, rgba(R,G,B,0.12) 3px,
  rgba(R,G,B,0.28) 3px, rgba(R,G,B,0.28) 8px);

/* 2. diag-med, medium diagonal (denser) */
background: repeating-linear-gradient(-45deg,
  rgba(R,G,B,0.11) 0px, rgba(R,G,B,0.11) 2px,
  rgba(R,G,B,0.27) 2px, rgba(R,G,B,0.27) 5.5px);

/* 3. diag-wide, bold diagonal bands */
background: repeating-linear-gradient(-45deg,
  rgba(R,G,B,0.08) 0px, rgba(R,G,B,0.08) 5px,
  rgba(R,G,B,0.26) 5px, rgba(R,G,B,0.26) 14px);

/* 4. dots, dense stipple */
background: radial-gradient(circle,
  rgba(R,G,B,0.4) 1.5px, rgba(R,G,B,0.1) 1.5px);
background-size: 6px 6px;

/* flat, solid fill (no pattern) */
background: rgba(R,G,B,0.28);

Wavy edge (optional)

Apply via clip-path on the fill element:

clip-path: polygon(
  0% 10%, 12% 5%, 25% 12%, 38% 3%, 50% 9%,
  62% 2%, 75% 11%, 88% 4%, 100% 8%,
  100% 100%, 0% 100%
);

When using wavy, add ~5% extra height to the fill to compensate for the wave.

Accent stripe

• 4px wide, left edge of cell/bar, full height
• Straight (not curved), clipped by parent’s border-radius via overflow: hidden
• Full opacity accent color

Footer

• Height: 48px, no HR/border above it
• Left: source label (9px uppercase, slate-600) + detail (8px, slate-700)
• Right: your site or handle (11px, slate-500) + your logo or headshot (52px circle, margin-top -8px, no border)

Logo/headshot embedding

Always base64-encode (cloud-synced file paths, e.g. iCloud or Google Drive, break in Playwright since it can’t authenticate to fetch them):

import base64
from pathlib import Path
logo_path = Path('/path/to/your/logo-or-headshot.png')
logo_b64 = base64.b64encode(logo_path.read_bytes()).decode()

Chart types

1. Horizontal bars

Best for: ranked metrics, 4-6 stats, “what’s missing” stories.

• Sentence-flow labels: read as continuation of the number, “85% don’t name a single machine they run”
• Number in separate column (110px, right-aligned), bar track fills remaining width
• Bar track: rgba(255,255,255,0.03), 6px border-radius
• Bar fill: reversed gradient (transparent to color), rounded right corners (border-radius: 0 6px 6px 0)
• 4px accent stripe on left edge of track
• Gaps: 2px between rows
• Padding: symmetric left and right

2. Stat grid (3x2 or 2x3)

Best for: 6 equal-weight stats as a scorecard.

• Grid of cells, 6px gaps, 8px border-radius
• Each cell: unfilled bg rgba(255,255,255,0.03), fill element inside (bottom-up, pattern or flat)
• Fill height = percentage value
• 4px accent stripe, left edge
• Number + tinted label anchored to bottom of cell
• Bottom-up fill preferred over left-right

3. Vertical columns (staircase)

Best for: showing relative scale, descending data.

• 6 columns, descending height = percentage
• Each column: full-height container with rgba(255,255,255,0.04) unfilled bg
• Fill element inside: height = percentage, pattern or flat
• Number + label inside the fill, anchored to top of fill
• No accent stripe on columns
• 6px gaps between columns, 8px border-radius

4. Newspaper (hero + strip)

Best for: one dominant stat + supporting context.

• Top: hero cell (bigger, ~1.2fr) with the headline stat, label, and subtitle
• Bottom: 5-column strip of supporting stats
• All fills bottom-up
• Bottom row: numbers baseline-aligned (position: absolute, fixed bottom offset) so labels don’t push numbers around

5. Quote + stat strip

Best for: provocative copy backed by data.

• Top: pull quote (Source Serif 4, 34px) with italic red emphasis word
• Bottom: 3-column stat strip with accent stripes and fills
• Quote mark: Source Serif 4, 56px, slate-800

6. Top 3 bold

Best for: 3 hero stats, no bar chart needed, maximum visual impact.

• 3 equal columns, each with massive number (90px), full sentence label, optional subtitle
• Vertical gradient fills (bottom-up)
• Accent stripes on left edge

Data sources

Your own data source

If you keep a local database with the stats you want to chart (SQLite, a CSV, an analytics export, whatever), query it directly rather than guessing at numbers. Adapt table and column names to your own schema. Example pattern using SQLite:

import sqlite3
conn = sqlite3.connect('/path/to/your/data.db')
cursor = conn.execute("SELECT column_a, column_b FROM your_table WHERE ...")

Flag for whoever adopts this skill: the original version of this pipeline step queried one specific private database with a hardcoded absolute path and a fixed table schema. That’s a real functional dependency, not just prose, so it can’t be copied as-is; wire it to wherever your own numbers actually live.

Manual data

Data provided directly in conversation: percentages, counts, labels.

Anti-patterns

• No matplotlib: HTML/CSS pipeline only
• No warm/artisanal aesthetics: professional services feel
• No “data science notebook” aesthetic: editorial/magazine
• No small labels: push toward “too big,” data IS the point
• No top-down fills: always bottom-up
• No soft/blurred fill edges: clean or wavy only
• No diagonal fill edges: wavy or flat only
• No footer HR: clean separation
• No headshot border: just shadow
• Don’t guess at data: query your data source or ask for the real numbers
• Don’t use local file paths for images: always base64 encode
• Don’t dump charts as terminal text: always open the image in a viewer

Design principles

1. Every pixel communicates: minimize whitespace, maximize information density
2. Grid adherence: structured layouts, consistent spacing, symmetric padding
3. Info hierarchy: title > numbers > labels > source. Each level visually distinct.
4. Define spaces, anchor elements: title centered in its space, padding matches on both sides
5. Sentence-flow labels: “85% don’t name their machines,” not “85% | No machines named”
6. Exaggerated scales: big fonts show font beauty, data IS the point
7. LinkedIn thumb test: must be readable and eye-catching at small size while scrolling
8. Fewer stats bigger: 3 huge stats beat 6 tiny stats for LinkedIn engagement

Quality checklist

• 1200x627px exact
• Title fills ~10/12 cols width (46px)
• Labels are BIG (18-19px) and tinted with parent hue
• Numbers are beautiful (Source Serif 4, 700 weight)
• Padding symmetric left/right
• Fills bottom-up, pattern only in filled area
• Accent stripes full-height, clipped by parent radius
• Footer: no HR, headshot has no border, 48px height
• Opened in an image viewer for a human to actually look at (not just read as terminal text)
• Visually verified via the Read tool before treating it as done